Biterms Topic Over Time:a Microblog Topic Evolution Analysis Model

Microblog text has the characteristics of short text,real-time user release and timestamp with Microblog system markup.Therefore,the sparsity of the short text and the relationship between text and time could directly affect the results of text mining in the process of Microblog text mining.In the process of modeling,the traditional topic model had ignored the sparsity of short text features and the dynamic evolution of text.In order to solve the problem,based on the topic model,Biterms Topic over Time(BTo T)topic model is proposed.The BTo T model introduces continuous time variables in the process of text generation,describing the topics evolution on the time dimension.In the document,the Biterms structure of topic sharing expands the characteristics of the essay and reduces the influence of feature sparsity on the mining effect of the topic model.Through the derivation of the Gibbs sampling method,three random parameter "document-topic","topic-word","topic – time" can be obtained in the BTo T model.Among them,the "topic-time" probability distribution represents the topic evolution process of corpus.The Microblog text with timestamp on Microblog platform is used as experimental data,and the appropriate model parameters is selected to verify the model.perplexity,topic similarity,classification accuracy and F1 value are used as evaluation coefficients,and Latent Dirichlet Allocation(LDA),Biterms Topic model(BTM)and Topic over Time(To T)are selected as contrast experiments.The experimental results show that the evaluation coefficient of the BTo T model is better than the contrast model,which can better accomplish the topic mining of Microblog text.It also analyzes the evolution process of the topic while ensuring the quality of the topic.